11.4 Vertical Pod Autoscaler (VPA)

⏱️ ~4 min read

TL;DR: VPA automatically adjusts a pod’s resource requests and limits based on observed actual usage. It solves the “what should I set my requests to?” question — especially useful for workloads with stable traffic patterns.

VPA vs HPA

How VPA Works

graph LR
    PODS[Running Pods\nActual usage data] -->|Collects metrics| REC[VPA Recommender\nML-based analysis]
    REC -->|Generates recommendation| VPA[VPA Object]
    VPA -->|Applies on pod restart\nor evicts to apply| POD[Pod with\nupdated requests]

VPA has three components:

• Recommender — analyzes historical usage and suggests optimal requests
• Updater — evicts pods that have outdated requests (to force a restart)
• Admission Plugin — mutates pod specs at creation time with VPA recommendations

VPA Modes

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: my-app-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: my-app

  updatePolicy:
    updateMode: "Auto"     # Options: Off | Initial | Recreate | Auto

  resourcePolicy:
    containerPolicies:
    - containerName: app
      minAllowed:
        cpu: "50m"
        memory: "64Mi"
      maxAllowed:
        cpu: "4"
        memory: "8Gi"

updateMode options:

💡 Tip: Start with updateMode: Off to see recommendations without disrupting running pods. Review for a few days, then switch to Initial for new pods, then Auto for full automation.

Reading VPA Recommendations

kubectl describe vpa my-app-vpa

Expected output:

Recommendation:
  Container Recommendations:
  - Container Name:  app
    Lower Bound:
      Cpu:     25m
      Memory:  64Mi
    Target:
      Cpu:     100m       ← VPA recommends setting requests to this
      Memory:  256Mi
    Uncapped Target:
      Cpu:     100m
      Memory:  256Mi
    Upper Bound:
      Cpu:     500m
      Memory:  512Mi

Use Target as your recommended requests values.

VPA on Minikube

VPA is not installed by default — it requires a separate installation:

# Clone the VPA repo and install (requires git)
git clone https://github.com/kubernetes/autoscaler.git
cd autoscaler/vertical-pod-autoscaler
./hack/vpa-up.sh

# Or just use updateMode: Off to see recommendations
# without installing the full VPA admission controller

On Minikube, VPA is primarily useful in Off mode for recommendations. Full Auto mode is better tested in cloud clusters.

VPA + HPA Together

Use VPA for memory (VPA recommends) and HPA for CPU scaling (HPA scales replicas):

# HPA on CPU only — let VPA handle memory right-sizing
spec:
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 60
  # Don't add memory HPA if VPA manages memory

# VPA on memory only — don't conflict with HPA's CPU scaling
spec:
  resourcePolicy:
    containerPolicies:
    - containerName: app
      controlledResources: ["memory"]   # Only manage memory, not CPU

Key Takeaways

✅ Quick Check

Q1: VPA recommends cpu: 100m for your pod which currently has cpu: 500m request. Will VPA immediately change the running pod?

Q2: Your app has a 5-minute startup time. VPA is in Auto mode and evicts the pod to update requests. What’s the impact?

Q3: Is there a way to use VPA without any pod disruption at all?